Deprecate non-future parallelism?

[kendonB]

There is duplication of ideas and effort in the multiple forms of parallelism supported by drake/drake-via-future, adding to (at least my) confusion over how to best use the package. I think it would be best to just use future for parallelism then have a different argument for transient/persistent/staged.

This requires (at least):

• Adding support for staged parallelism using future.
• Adding a future evaluator for using Makefile parallelism. i.e. future::plan(makefile) OR deprecating Makefile parallelism.
• Adding deprecation messages in the relevant functions

Also mentioned here but didn't get it's own issue: #227

[wlandau]

It's a good thought, but I do not think it is time to deprecate any parallel drake is still in a brainstorming phase when it comes to HPC. New paths and technologies are emerging fast, and it is not yet clear where drake will settle. Also, in future-based parallelism specifically, there is extra overhead to make sure workers can be deployed remotely. This does not bode well for imports or large numbers of quick targets.

[wlandau]

Lastly, I have actually been trying to get rid of staged parallelism. I could not quite do it for the last CRAN release because there are some projects that benefit (e.g. #369 (comment)), and for projects that require remote workers, I believe transient workers ("future" parallelism) is uniformly better.